利奈唑胺TDM超阈值风险预测模型的建立与评估

Establishment and evaluation of a risk prediction model for linaclotide TDM exceeding the threshold

关键词:  利奈唑胺  治疗药物监测  超阈值  风险预测模型  列线图

Key words:  Linezolid  Therapeutic drug monitoring  Exceeding the threshold  Risk prediction model  nomogram

摘要:

目的 研究利奈唑胺治疗药物监测(TDM)超阈值的危险因素并构建可视化的超阈值风险预测模型,进以优化临床利奈唑胺个体化用药。方法 根据纳排标准,筛选厦门大学附属中山医院2023年6月至2024年10月使用利奈唑胺并对其行TDM的住院患者,使用高效液相色谱法检测血清药物浓度,将浓度范围在2-8 mg.L_-1^的患者列为达标组,≥8 mg.L_-1^的患者列为超阈值组,收集两组患者的临床资料(包括年龄、性别、身高、体重、基础疾病、用药情况、相关支持治疗、炎症指标、血常规、肝肾功能等),使用单因素分析和多因素分析筛选识别特征预测变量并构建列线图模型,使用受试者工作特征(ROC)曲线、校准曲线、决策曲线和临床影响曲线对列线图模型进行评估。结果 本研究共纳入了180例患者(282次TDM),达标组153次TDM(男109次,女44次),超阈值组129次TDM(男87次,女42次)。多因素分析显示年龄(OR=1.052,95%CI:1.026, 1.078)、血红蛋白(OR=0.965,95%CI:0.934, 0.998)、低蛋白血症(OR=2.440,95%CI:1.192, 4.996)为利奈唑胺TDM超阈值的独立危险因素。根据上述独立危险因素构建列线图模型,结果显示ROC曲线的AUC为0.7721(95%CI:0.718-0.8263)。校准曲线结果显示模型预测值与实际结果拟合度良好(Bootstrap自抽样法重复抽样1000次,平均绝对误差为0.015),表明列线图模型的良好可靠性和特异性。模型决策曲线DCA结果显示在风险阈值约有0.08-0.95概率范围内预测利奈唑胺TDM超阈值具有较高的净获益。结论 本研究构建的利奈唑胺TDM超阈值风险预测模型具有良好的可靠性和临床适用性,可帮助临床快速识别风险患者,为临床快速制定利奈唑胺个体化给药方案提供科学依据。

Abstract:

ABSTRACT: OBJECTIVE To investigate the risk factors for exceeding the threshold of therapeutic drug monitoring (TDM) with linezolid and construct a visual model for predicting the risk of exceeding the threshold, in order to optimize personalized clinical use of linezolid. METHODS According to the inclusion and exclusion criteria, hospitalized patients treated with linezolid at Zhongshan Hospital affiliated with Xiamen University from June 2023 to October 2024 were screened and subjected to TDM. The serum drug concentration was detected using high-performance liquid chromatography. Patients with a concentration range of 2-8 mg·L-1 were classified as the standard group, and patients with a concentration of ≥ 8 mg·L-1 were classified as the over threshold group. Clinical data of the two groups of patients were collected, including age, gender, height, weight, underlying diseases, medication use, relevant supportive treatment, inflammatory indicators, blood routine, liver and kidney function, etc. Univariate analysis and multivariate analysis were used to screen and identify feature predictive variables, and a column chart model was constructed. Receiver operating characteristic (ROC) curves and calibration were used to identify the predictive variables. Evaluate the column chart model using curves, decision curves, and clinical impact curves. RESULTS a total of 180 patients (282 TDMs) were included in this study. The standard group had 153 TDMs (109 for males and 44 for females), while the over threshold group had 129 TDMs (87 for males and 42 for females). Multivariate analysis showed that age (OR=1.052, 95% CI: 1.026, 1.078), hemoglobin (OR=0.965, 95% CI: 0.934, 0.998), and hypoalbuminemia (OR=2.440, 95% CI: 1.192, 4.996) were independent risk factors for exceeding the TDM threshold of linezolid. Based on the independent risk factors mentioned above, a column chart model was constructed, and the results showed that the AUC of the ROC curve was 0.7721 (95% CI: 0.718-0.8263). The calibration curve results show that the predicted values of the model fit well with the actual results (Bootstrap self sampling method repeated 1000 times, with an average absolute error of 0.015), indicating the good reliability and specificity of the column chart model. The DCA results of the model decision curve show that within a probability range of approximately 0.08-0.95 at the risk threshold, predicting linezolid TDM exceeding the threshold has a higher net benefit. CONCLUSION The TDM over threshold risk prediction model for linezolid constructed in this study has good reliability and clinical applicability, which can help quickly identify high-risk patients in clinical practice and provide scientific basis for the rapid development of individualized linezolid administration plans. KEYWORDS: Linezolid; Therapeutic drug monitoring; Exceeding the threshold; Risk prediction model; nomogram